Handling Concept Drift in Data Streams by Using Drift Detection Methods

The growth and development of the information and communication technology of the present era resulted in huge amount data generation. It is found that the rate of data distribution is very high. The data which is generated with varying distributions is referred to as data stream. Few examples to quote, data generated with regard to applications related to mobile networks, sensor networks, network traffic monitoring and network traffic management, etc. It is found that, the data generation process often change with respect to data distribution for any kind of concept, i.e. application which is referred to as concept drift. Handling concept drift is a challenging task. It is impossible to develop a model as it will be inconsistent in nature because of continuous change. The present work emphasises on handling the concept drifts, using different drift detection methods using Massive Online Analysis Framework. The important feature of the present study is varying size of a data stream (50,000–250,000). Totally the Concept Drift is handled using 11 drift detection methods using 2 stream generators abrupt and gradual under this frame work respectively.